"Get Jenkins CLI" failes with a 503 from localhost

[sumdog]

So I'm using this plugin via Vagrant (created an ansible/roles directory; my .yml that includes the ansible-jenkins role is in ansible. Vagrant file is in .)

I see the following:

TASK: [ansible-jenkins | Get Jenkins CLI] ************************************* 
failed: [build] => {"dest": "/opt/jenkins/jenkins-cli.jar", "failed": true, "response": "HTTP Error 503: Service Unavailable", "state": "absent", "status_code": 503, "url": "http://localhost:8080/jnlpJars/jenkins-cli.jar"}
msg: Request failed

Seems to be failing here where you get the CLI jar file from the running Jenkins instance:

# Get Jenkins CLI from localhost
- name: Get Jenkins CLI
  sudo: yes
  action: get_url url=http://localhost:8080/jnlpJars/jenkins-cli.jar dest={{ jenkins.cli_dest }} mode=0440

Not sure if something is different with the new Jenkins version... leaving work right now but I'll try to figure this out when I get home and get up a patch.

[sumdog]

I think Jenkins just wasn't online yet. A second call to provision works. Doesn't look like anisble's get_url has a retry/timeout option either.

[danvaida]

@sumdog see line 18 in my PR:
danvaida@1219ad0

[sumdog]

I tried the @danvaida pull request references above, but even if I set the timeout to 50000, I still get the 503; service unavailable. A second provision works fine.

I have a feeling ansible needs a retry option for the get_url for this to work sanely. Either that or we need to consider shelling out to wget.

[danvaida]

@sumdog What version of Ansible are you using? Have you experienced any network quirks with the target box? Works totally fine for me. Shelling it out is not so "ansiblish" :)

[sumdog]

@danvaida I got the 503 consistently on both my work machine (i7, 16GB ram, SSD) and my home laptop (i5, 8GB ram). I my case the timeout=xx that didn't fix the issue either. I think timeout would work if it were simply waiting, but I'm actually getting a 503 returned.

I have a feeling it's simply Jenkins not coming up fast enough in my VM. A second request always works. In the following commit I have an ugly hack that works:

BigSense/vSense@e85891a

I'm not going to issue a pull request because it is pretty ugly. I'll leave this comment here through in case anyone else is interested.

This might not be fixable unless a feature request is put in to Ansible to add a retry to get_url that can also deal with retrying on 5xx responses.

[danvaida]

@sumdog Not necessarily :)
I'll modify my fork tomorrow as follows:

 register: get_url_result
 until: "'OK' in get_url_result.msg"
 retries: 5
 delay: 10

There. That should fix this corner case.

LE: see respective lines in my commit: danvaida@dbbb8db